Electronic device and method with image encoding and decoding
A processor-implemented method includes determining weights of a plurality of channel blocks comprised in a neural network to encode an image, selecting one or more channel blocks from the plurality of channel blocks to encode the image based on the weights and a compression rate of the image, and generating an encoded image by encoding the image using a sub-neural network comprising channels comprised in the one or more channel blocks.
1 . A processor-implemented method, the method comprising:
determining weights of a plurality of channel blocks comprised in a neural network, each of the plurality of channel blocks includes a predetermined number of channels;
selecting one or more channel blocks from the plurality of channel blocks based on the weights and a compression rate of an image; and
generating an encoded image by encoding the image using a sub-neural network comprising channels comprised in the selected one or more channel blocks.
2 . The method of claim 1 , wherein the determining of the weights of the plurality of channel blocks comprised in the neural network comprises inputting the image to a previously trained channel block prediction network, and
wherein the generating the encoded image by encoding the image is implemented using only the selected one or more channel blocks of the plurality of channel blocks.
3 . The method of claim 2 , wherein output channels of a last layer comprised in the channel block prediction network respectively correspond to the plurality of channel blocks comprised in the neural network.
4 . The method of claim 1 , wherein the selecting of the one or more channel blocks from the plurality of channel blocks comprises:
determining a number of the one or more channel blocks based on the compression rate; and
selecting the one or more channel blocks from the plurality of channel blocks as much as the number, based on the weights.
5 . The method of claim 4 , wherein the determining of the number of the one or more channel blocks based on the compression rate comprises determining the number using a number of channels corresponding to the compression rate and a number of channels comprised in the plurality of channel blocks.
6 . The method of claim 4 , wherein the selecting of the one or more channel blocks as much as the number comprises selecting from the plurality of channel blocks as much as the number in order of highest corresponding weight.
7 . The method of claim 1 , wherein the generating of the encoded image comprises:
extracting feature information of the image through the sub-neural network; and
generating the encoded image by encoding the feature information.
8 . The method of claim 7 , wherein the extracting of the feature information of the image comprises:
extracting a latent variable of the image from the image;
extracting side information on the image based on the latent variable; and
determining a Gaussian distribution parameter value of the latent variable based on the side information.
9 . The method of claim 8 , wherein the generating of the encoded image comprises:
determining encoded side information by encoding the side information;
determining an encoded latent variable by encoding the latent variable based on the Gaussian distribution parameter value; and
generating the encoded image comprising the encoded side information and the encoded latent variable.
10 . The method of claim 1 , further comprising adding information indicating the one or more channel blocks to the encoded image to decode the encoded image.
11 . A non-transitory computer-readable storage medium storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the method of claim 1 .
12 . A processor-implemented method, the method comprising:
selecting one or more channel blocks from a plurality channel blocks of a neural network based on information indicating one or more channel blocks, each of the plurality of channel blocks includes a predetermined number of channels; and
generating a reconstructed image by decoding an encoded image using a sub-neural network comprising channels comprised in the selected one or more channel blocks.
13 . The method of claim 12 , wherein the encoded image comprises encoded side information obtained by encoding side information extracted based on an encoded latent variable and side information extracted based on the latent variable wherein the encoded latent variable is obtained by encoding a latent variable extracted from the image.
14 . An electronic device comprising:
one or more processors;
a memory comprising one or more non-transitory storage media that store instructions that, when executed by the one or more processors, configures the electronic device to:
determine weights of a plurality of channel blocks comprised in a neural network, each of the plurality of channel blocks includes a predetermined number of channels;
select one or more channel blocks from the plurality of channel blocks based on the weight and a compression rate of an image; and
generate an encoded image by encoding the image using a sub-neural network comprising channels comprised in the selected one or more channel blocks.
15 . The electronic device of claim 14 , wherein, for the generating of the encoded image, the execution of the instructions further configures the electronic device to:
extract feature information of the image through the sub-neural network using only the selected one or more channel blocks of the plurality of channel blocks; and
generate the encoded image by encoding the feature information.
16 . The electronic device of claim 15 , wherein the execution of the instructions further configures the electronic device to:
extract a latent variable for the image, extract side information on the image based on the latent variable; and
determine a Gaussian distribution parameter value of the latent variable based on the side information.
17 . The electronic device of claim 16 , wherein the execution of the instructions further configures the electronic device to:
determine encoded side information by encoding the side information;
determine an encoded latent variable by encoding the latent variable based on the Gaussian distribution parameter value; and
generate the encoded image comprising the encoded side information and the encoded latent variable.
18 . The electronic device of claim 14 , wherein the execution of the instructions further configures the electronic device to add information indicating the one or more channel blocks to the encoded image to decode the encoded image.
19 . An electronic device comprising:
one or more processors; and
a memory comprising one or more non-transitory storage media that store instructions that, when executed by the one or more processors, configures the electronic device to:
select one or more channel blocks from a plurality channel blocks of a neural network based on information indicating the one or more channel blocks, each of the plurality of channel blocks includes a predetermined number of channels; and
generate a reconstructed image by decoding an encoded image using a sub-neural network comprising channels comprised in the selected one or more channel blocks.
20 . The electronic device of claim 19 , wherein the encoded image comprises encoded side information obtained by encoding side information extracted based on an encoded latent variable and side information extracted based on the latent variable, and
wherein the encoded latent variable is obtained by encoding a latent variable extracted from the image.